Federal regulation 49 CFR 193 and standard NFPA 59A require the use of validated consequence models to determine the vapor cloud dispersion exclusion zones for accidental liquefied natural gas (LNG) releases. For modeling purposes, the physical process of dispersion of LNG release can be simply divided into two stages: source term and atmospheric dispersion. The former stage occurs immediately following the release where the behavior of fluids (LNG and its vapor) is mainly controlled by release conditions. After this initial stage, the atmosphere would increasingly dominate the vapor dispersion behavior until it completely dissipates. In this work, these two stages are modeled separately by a source term model and a dispersion model due to the different parameters used to describe the physical process at each stage. The principal focus of the source term study was on LNG underwater release, since there has been far less research conducted in developing and testing models for the source of LNG release underwater compared to that for LNG release onto land or water. An underwater LNG release test was carried out to understand the phenomena that occur when LNG is released underwater and to determine the characteristics of pool formation and the vapor cloud generated by the vaporization of LNG underwater. A mathematical model was used and validated against test data to calculate the temperature of the vapor emanating from the water surface. This work used the ANSYS CFX, a general-purpose computational fluid dynamics (CFD) package, to model LNG vapor dispersion in the atmosphere. The main advantages of CFD codes are that they have the capability of defining flow physics and allowing for the representation of complex geometry and its effects on vapor dispersion. Discussed are important parameters that are essential inputs to the ANSYS CFX simulations, including the mesh size and shape, atmospheric conditions, turbulence from the source term, ground surface roughness height, and effects of obstacles. A sensitivity analysis was conducted to illustrate the impact of key parameters on the accuracy of simulation results. In addition, a series of medium-scale LNG spill tests have been performed at the Brayton Fire Training Field (BFTF), College Station, TX. The objectives of these tests were to study key parameters of modeling the physical process of LNG vapor dispersion and collect data for validating the ANSYS CFX prediction results. A comparison of test data with simulation results demonstrated that CFX described the physical behavior of LNG vapor dispersion well, and its prediction results of distances to the half lower flammable limit were in good agreement with the test data.